The reviewed record of science sign in
Pith

arxiv: 2506.03637 · v2 · pith:ZZ3KX72R · submitted 2025-06-04 · cs.CL · cs.AI· cs.LG

RewardAnything: Generalizable Principle-Following Reward Models

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:ZZ3KX72Rrecord.jsonopen to challenge →

classification cs.CL cs.AIcs.LG
keywords principlesrewardlanguagemodelsrewardanythingnaturalpreferencerabench
0
0 comments X
read the original abstract

Reward Models, essential for guiding Large Language Model optimization, are typically trained on fixed preference datasets, resulting in rigid alignment to single, implicit preference distributions. This prevents adaptation to diverse real-world needs-from conciseness in one task to detailed explanations in another. The standard practice of collecting task-specific preference data and retraining reward models is resource-intensive, often producing biased rewards, and limits practical application. We introduce generalizable, principle-following reward models. We propose that RMs should understand and adhere to dynamically provided natural language specifications of reward principles, similar to instruction-following in LLMs. To measure this capability, we develop RABench, a comprehensive benchmark for RMs focusing on generalization across diverse principles. Evaluations on RABench reveal poor generalization of current RMs. As a solution, we present RewardAnything, a novel RM designed and trained to explicitly follow natural language principles. We achieve SotA performance with RewardAnything in traditional RM benchmark simply by specifying a well-defined principle, and results on RABench show we excel in adapting to novel principles without retraining. Furthermore, RewardAnything integrates seamlessly with existing RLHF methods and we show by a case study on how to automatically and efficiently align LLMs with only natural language principles.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Support Vector Rubrics: Closing the Gap Between Self-Generated and Human Rubrics

    cs.CL 2026-06 unverdicted novelty 7.0

    SVR learns a bank of contrastive rubrics from preference data via max-margin boundaries and prompt-conditioned selection, narrowing the gap to human rubrics on RubricBench from 24.1 to 0.3 points.

  2. DoGMaTiQ: Automated Generation of Question-and-Answer Nuggets for Report Evaluation

    cs.CL 2026-05 unverdicted novelty 7.0

    DoGMaTiQ automates creation of QA-based nuggets via document-grounded generation, paraphrase clustering, and quality subselection, producing report evaluations that correlate strongly with human judgments on cross-lin...

  3. Beyond Verifiable Rewards: Rubric-Based GRM for Reinforced Fine-Tuning SWE Agents

    cs.LG 2026-03 unverdicted novelty 7.0

    A rubric-based generative reward model improves reinforced fine-tuning of SWE agents by supplying richer behavioral guidance than binary terminal rewards alone.

  4. Skill-RM: Unifying Heterogeneous Evaluation Criteria via Agent Skill

    cs.LG 2026-06 unverdicted novelty 6.0

    Skill-RM unifies heterogeneous reward criteria by modeling reward computation as dynamic execution of a reusable Reward-Evaluation Skill within an agent framework.

  5. DoGMaTiQ: Automated Generation of Question-and-Answer Nuggets for Report Evaluation

    cs.CL 2026-05 unverdicted novelty 6.0

    DoGMaTiQ automates QA-nugget creation via document-grounded generation, paraphrase clustering, and quality-based subselection, yielding strong rank correlations with human judgments on cross-lingual TREC tasks.

  6. RLBFF: Binary Flexible Feedback to bridge between Human Feedback & Verifiable Rewards

    cs.CL 2025-09 unverdicted novelty 6.0

    RLBFF extracts binary principles from human feedback to train reward models that outperform Bradley-Terry models on RM-Bench and JudgeBench and enable customizable inference-time focus for LLM alignment.

  7. Beyond Overlap Metrics: Rewarding Reasoning and Preferences for Faithful Multi-Role Dialogue Summarization

    cs.CL 2026-04 unverdicted novelty 5.0

    A reasoning-distillation plus dual-reward GRPO method for multi-role dialogue summarization matches ROUGE and BERTScore baselines while improving factual faithfulness and preference alignment on CSDS and SAMSum.